3750 stack master failure

[G.I II]

We have a stack of a 3750E with a 3750X, when we turn off the master, a disconnection occurs for around one minute while only one request timeout is seen when the slave is turned off. Is this normal?

Details: in other words, when the slave switch is elected as the new master all the port-channels and some physical ports become in 'down' state before reobtaining the up state again, this results in a one minute outage. The question is: is it normal to have a layer2 control traffic interruption when the master fails and re-election occurs and if yes, what is the normal downtime in seconds. the layer2 control information such as port-channels LACP protocol are not synchronised to all slave switches by the master?

[Reza Sharifi]

One minute between failure is too long.  Here are some numbers from Cisco's documentation:

On stack master failure, another switch in the stack takes over

1:N master redundancy

Reconvergence times tested under heavy load:

Layer 1 failure is detected in several microseconds

Layer 2 failure ~ mseconds

Layer 3 link failure—sub 200 mseconds

Layer 3 member failure—sub 300 mseconds

Layer 3 master failure—up to eight seconds

HTH

[G.I II]

Hi Reza,

Layer 2 failure ~ mseconds is when the master switch fails?

BR

[Reza Sharifi]

Correct. This is when the master fails.

[G.I II]

ok, do you think that 'stack-mac persistent timer 0' command which is still not implemented will help in decreasing the downtime from 1 minute to mseconds or it will just decrease the time needed to reach the management IP again?

[Reza Sharifi]

I know enabling it defiantly helps when the devices are running layer-3 routing protocols like OSPF

I know your devices are layer-2, but I would try anyway

please keep us posted on the results.

HTH

[G.I II]

i think it should help also for layer2, this is what i found in my research :

With LACP, the system-id uses the stack MAC address from the stack master, and if the stack master changes, the LACP system-id can change. If the LACP system-id changes, the entire EtherChannel will flap, and there will be an STP reconvergence. Taken from the following reference:

http://www.cisco.com/en/US/docs/switches/lan/catalyst3750x_3560x/software/release/15.0_1_se/configuration/guide/3750X_scg.pdf

I will update u after doing the test

many thanks

[G.I II]

test done and what was mentioned in the previous post resolved the problem